recent investigations of intelligence and its measurement

RECENT INVESTIGATIONS OFINTELLIGENCE AND ITS MEASUREMENT*

By PHILIP E. VERNON, M.A., Ph.D.EARLY eugenic investigations of mental

qualities were based, for the most part,on the investigators' own assessments

of intellectual eminence or special talent,social worth or depravity, and the like.Obviously the subjectivity of these judg-ments was open to criticism, however fair-minded the investigator. Hence the inven-tion of objective tests of intelligence wasreceived with enthusiasm. They appeared tooffer scope for far more scientific studies ofinheritance. Thus the Eugenics Society hassponsored, or has undertaken the publica-tion of, numerous and outstanding researchesby psychologists such as Burt, Thomson,R. B. Cattell, Fraser Roberts and Terman,which were based on intelligence tests. Itwas of course realized that tests either of theBinet type or the group verbal type (e.g.Otis) were not pure measures of innateability. Early studies like those of Gordonwith canal-boat children and Freeman withAmerican foster-children revealed a distinctinfluence of education and cultural back-ground. This was confirmed by Newman,Freeman and Holzinger's work on identicaltwins brought up apart. However, analysesof foster-children and twin data by Burksand others indicated that environmentaleffects were unlikely to alter an IntelligenceQuotient by more than ten points up ordown; or, more precisely, that the contribu-tion of heredity is three or four times asgreat as that of environment in determiningindividual differences in intelligence. Thusmost psychologists, in this country at least,who were interested in eugenic problemswere satisfied that ordinary intelligence testswere adequate tools for genetic investigations.Moreover, when Cattell (7) employed non-verbal group tests, based on pictorial ordiagrammatic material, which would pre-sumably be less susceptible to educational

* A paper read before the E-ugenics Society on May29th, I95I.

influence, he obtained much the same nega-tive correlation with family size as didworkers with verbal tests.

Nevertheless, recent investigations bymyself and others have forced me to theconclusion that, while intelligence tests areadmirable instruments for practical purposessuch as educational and occupational selec-tion and guidance within any one culturalgroup, they cannot be regarded as sufficientlypure measures of innate ability to beemployed in comparisons between differentgroups such as races or nations, nor forgenetic studies. And this discouraging con-clusion applies to non-verbal, performance,or other tests as much as to verbal Binet orgroup tests. My views have altered sinceI949 (27), largely because two crucialeugenic studies-those of the Scottish MentalSurvey (2I) and of Cattell (8)-have givennegative results. If the tests that theyemployed were adequate measures of inbornintelligence, then I do not see how they couldfail to disclose some decline over the pastfifteen years. Thomson (24) and Cattell arecommendably cautious in their discussions ofpossible reasons for the rises in their grouptest means, and the lack of any drop onindividual tests. But their explorations ofpossible genetic factors seem to lead only toblind alleys, and even Penrose's (I9) ingeni-ous picture of heterosis could hardly com-pensate for the relatively large fall that wasanticipated. Thus they appear to incline tothe view that practice or sophisticationeffects may have distorted the results of thegroup tests, particularly among girls, andthat improvements in health and educationmay have played some part ; while Thomsonadmits the possibility that the negativerelation between intelligence scores andfamily size may be partly environmental.Most of my paper to-day, therefore, will bedevoted to relevant studies of practice andenvironmental effects, and their implications.

B I S,

I26 THE EUGENICS REVIEW

First, however, I would like to giveyou the preliminary results of a large-scalesurvey of the abilities of some 9,ooo Armyrecruits. When the Scottish results werepublished I suggested that part of the diffi-culty of interpretation arose from using suchfactorially complex tests as Moray Houseand Stanford-Binet, and that further datashould be obtained with the aid of factoranalysis to supply relatively pure measuresof Spearman's g.

A Survey of Army RecruitsThe scores of about IO,OOO male National

Service recruits, mostly aged i8, wereobtained on six mental tests. Their heightand their physical and eyesight gradings onthe Pulheems system were available, togetherwith records of the county in which theylived. During an interview with a personnelselection officer they were classified by theplace they had chiefly lived as rural versusurban and as native or migrant; finally thenumbers of living siblings were ascertained.Such recruits provide a fairly representativesample of the population, but unfortunatelyfew agricultural workers were being called upat the time (the end of I947). Also there wasa wastage of 6 per cent of men who failed tosupply the interview data, and these weresomewhat inferior in intelligence.* Thematerial was collected by the personnelselection staff of the War Office's Directorateof Manpower Planning, and all the essentialtabulations and calculations were done fromHollerith cards by the Statistics Branch ofthe Admiralty. I am most grateful to thoseconcerned and to my colleague in the SeniorPsychologist's Department of the Adniralty-Mr. E. Elliott-for their assistance. Afuller account than I can give to-night willbe published in due course.

Table I shows the best solution I have beenable to reach so far to the analysis of theinter-correlations; it gives the loadings orsaturations of the tests with four mainfactors or underlying types of ability: g

* There were 603 such cases, whose mean intelligence" quotient," on the scale used below, was 96-5. Anumber of others had failed to take one or more tests,leaving a total of 9,183.

general, verbal-educational, spatial-mechani-cal and physical. The results are fairlylogical, but one should note the overlappingbetween the g and the education factors. Notonly does an arithmetic test show the highestg-loading but also a non-verbal test and amechanical one have considerable educa-tional content. Natives, you will see, areslightly (but significantly) inferior to

TABLE ILOADINGS oF TESTS AND OTHER MEASURES ON FOURFACTORS EXTRACTED AND ROTATED BY THE CENTROID

METHODg v: ed mech. phys.

Arithmetic-maths.... .....73 '57Verbal ability ... ... *6i '65Clerical ... ... ... *71 *56 o7Education standard ... 6o *71 *o6Non-verbal intelligence ... *72 .40 '20Mechanical problems .. .54 50 *47 *12Mechanical assembly ... .31 *45 *i6P (physical) ... ... *o8 *37EE (eyesight) ... ... *01 -23 .32Height ... ... ... '09 '33 *13 '20

Variance, per cent ... 26'8 2I-4 4.8 3'2

Native ... ... -04 -'10 -07 -04Urban ... ... ... -'04 '01 -'03 -IIISiblings ... ... ... -'34 -'I7 -'09 -'04

migrants on all factors. Urban dwellers seemto be slightly inferior in all respects, excepteducation, and particularly in physicalqualities. But I am afraid that the lack ofagricultural workers may have affected thesefigures; we shall see later that the relation-ship is not linear. As regards family size, itis quite clear that the negative correlationexists not so much with educafion as withour general factor. But as the general factoritself is somewhat mixed up with educationthis result does not carry us much further.It is interesting that the correlation of-.34 is higher than is usually found amongchildren; presumably this is due to morefamilies being complete when a cross-sectionis taken at i8 than at ii years.

I had hoped to calculate independentfactor scores, but the labour proved too great,and instead I merely combined the results ofcertain tests which were most highly loadedwith each factor. For ease of comprehensionI converted these combined scores intostandard scores with a mean of IOO and a

RECENT INVESTIGATIONS OF INTELLIGENCE AND ITS MEASUREMENT I27

standard deviation of i6, so that they shouldall be comparable to I.Q.s.* Table II showsthe mean standard scores for g, education,mechanical and physical tests for each familysize. Note, however, that because of themethod of derivation the education andmechanical scores are considerably impreg-nated with g. You will see that the decline

TABLE IIMEAN STANDARD SCORES OF RECRUITS WITH GIVENNUMBERS OF SIBLINGS ON TESTS OF DIFFERENT TYPESNo. of Freq. Mean Scores Scottish

Siblings % g ed. mec. phys. MentalSurvey

0 I3.3 io6-6 IO7-2 IO4-6 IO2-3 105.3I 220 IO5-8 105.8 I04-3 ioir6 105*I2 i8*8 ioi-8 IOI*7 IOI-7 IOO18 ioI.63 13-8 98-8 98-5 99-5 99-8 98-64 IO4 94-9 94-7 96.7 98.7 95.85 7-8 93-2 92-9 95-2 97.7 94-26 52 92-4 92-6 93-5 96*4 92-87-8 5.5 88-9 9I-6 92-9 96.5 9I-89-II 2-8 87.9 88-2 9o-6 96-2 9O-I12-17 0.4 87.2 86z2 gi-6 95.8 86.5

is almost identical on the more g-loaded andthe more educational tests, and that bothclosely parallel the results of the ScottishMental Survey. On more mechanical teststhe decline is less marked, and it is quitesmall on physical measures; indeed there isscarcely any drop beyond six siblings.

Turning to geographical differences: thecounties were grouped into twelve majorregions, which were so chosen that no regioncontained less than 5 per cent of the total,and each would be fairly homogeneous inrespect of industrialization. Table III showsthat variations in g scores in differentregions are remarkably large, ranging froman average of I04 in the Home Counties toninety-four in South-West Scotland. Wales,Lancashire, Stafford and Warwickshire aredistinctly below average-ninety-seven- toninety-eight; but the whole of the rest ofEngland and Scotland, including London, isaround ioo -to io2. There were, of course,variations between counties within regions;but numbers were too small for these to behighly significant. Surprisingly the regionaldifferences on educational tests are smaller

* Such units are not normally employed by Servicepsychologists, who are well aware of the misinterpreta-tions to which I.Q.s, particularly among adults, areliable.

than on g tests, while mechanical differencesare largest of all. The South-West Scots andWelsh are relatively well educated, the EastAnglians are significantly poorer than wouldbe expected from their intelligence. Onmechanical tests it is the less industrializedand more healthy regions that do best,though Warwick-Stafford does pull upslightly. The physical results run almostparallel to the mechanical and are incident-ally very similar to those discovered by W. J.Martin (i6) (in so far as our differentlygrouped regions can be compared.)

TABLE IIIMEAN STANDARD SCORES OF RECRUITS FROM DIFFERENT

REGIONSRegion of

Great Britain Nos. g ed. mec. phys,

Home Counties ... i,663 104.0 103.0 I04.3 I03.0East Anglia, Beds,Cambs, Northants 474 IOI*7 IOO1O 103.0 ioo-6

S.W. and Hants ... 563 IOI5 IOI'0 103.7 102.5Berks, Bucks, Oxon,

Glos, Here, Worcs 540 IOI-4 I00.9 I03-2 IO2*7Leics, Notts, Derby,Yorks W., Ches I,128 101-4 IOI*0 100.3 99.7

London ... ... 695 ioo*6 99.9 ioo.8 99*8Yorks E. & N.,Northumb, Dur,Cumb ... ... 502 99g6 ioo0o 99-I 99*2

Scotland E. and N.Counties ... ... 486 99'6 99-4 97.0 98-9

Wales ... ... 466 97.9 IOO-O 97-0 97-2Lancashire ... ... 1,125 97-2 97-8 96*5 97-6Warwick, Staffs,

Salop ... ... I,O46 97.1 96-9 98-4 99.5Glasgow and S.W.

Scotland ... ... 495 93.7 97.o 91.4 95-0

Natives ... ... 74p8 99-0 99 0 99-2 99.5Migrants ... ... 25*2 I02*8 102*9 I02-4 Ioi*6

Distancefromopen

countrymiles %

Rural ... 0 4.9 I00.4 99-8 IO2-5 102.8Village or

small town < I 22-4 99'8 IOO1O 1004 101.0Middle ofmod. town I-2 i886 Iroo9 100.7 IOO2 IOO2

Edge of largetown ... 1-2 15.0 I03-0 IO3I 102-4 100.7.

Middle oflargetown 3+ 39'I 98-4 98-5 98.5 98-7Migrants are uniformly superior to natives.

Note that they constitute only a quarter ofthe population in spite of all the war-timemigrations. Martin found 20 per cent in

i28 THE EUGENICS REVIEW

I939. My rural-urban classification refers tothe place in which the recruit has chieflylived (not necessarily where he was born norwhere living at present), and was based onthe distance of that place from open country.Thus the first group represents those who saythey live surrounded by -open country,the second those living in small towns orvillages where open country can be seen fromclose by their homes. You will see that thesize of the town is not crucial. Those livingon the outskirts of large towns are highestand those in the centre lowest on almost allpsychological tests. The true countrydwellers do not appear to be inferior inabilities; they are actually highest onmechanical as well as physical measures.While I have admitted a possible lack ofagricultural workers in my sample, the factthat 5 per cent of the total fall in the mostrural group and 22 per cent in the village-small town group suggests that the loss isslight. Mrs. Bosanquet's (2) careful surveyshowed, like mine*the physical superiority ofrural populations. Her evidence for intel-lectual inferiority was by no means unanim-ous, and she wondered whether such inferi-ority as was found might not be ascribed tothe content of the tests favouring urbanchildren. I would add to this the suggestionthat urban children are more likely thanrural to be practised, or sophisticated, intaking tests; and we shall see below thatsuch practice effects may be very consider-able. To conclude then: there is too muchdoubt about the representativeness of mysample for me to contradict the commonbelief in the lower intelligence of the ruralpopulation. But at least it would appearthat any inferiority lies more in educationallevel than in general or practical ability.The more important finding of my survey,from the eugenic standpoint, is that the bestmeasure of intelligence that I have been ableto extract from six mental tests behaves injust the same way as tests with obviouseducational content in predicting a decline innational intelligence-a decline which has sofar not been borne out by direct evidence.

Practice and Coaching Effects onIntelligence TestsTurning then to environmental effects on

test scores: these may be classified underfive headings-

i. Coaching or teaching the right answerson the actual test used.

2. Practice in taking the test itself.3. Practice on other similar tests.4. Coaching on other similar tests.5. More generalized educational training

or cultural influences which are inci-dentally reflected in test scores.

The first-coaching on the test itself-isobviously possible, but is unimportant be-cause care is normally taken nowadays toprevent leakages. Secondly-practice effectsfrom repeating the same test; these wereshown to be so large in the Army during thewar that additional norms had to be pro-vided when men were retested. The improve-ments ranged from 2.i per cent to 8.6 percent on different tests, in terms of I.Q. unitswith a standard deviation of fifteen. Theywere highest among tests with unfamiliarcontent or with elaborate instructions andrestricted time limits, such as ProgressiveMatrices and a clerical test; least markedamong straightforward tests such as arith-metic, where also the timing was generous.More recently Dr. Heim (6) and her col-laborators have shown that repeated practiceat a timed group test (having adequateceiling) produces almost indefinite rises,though these tend to tail off after about thefifth retest. Obviously this is a quite artifi-cial situation; practice or coaching on othersimilar tests is of far greater importance. Itis particularly serious now that intelligencetests are used by almost all educationauthorities for competitive entry to grammarschools at eleven years. In many primaryschools, much of the final year is spent incoaching children to pass intelligence andobjective English and arithmetic tests, inspite of the efforts of the authorities toprevent it. Not only is this unfair to theschools which do refrain from coaching butalso it naturally renders the original testnorms valueless and it distorts the results of


any comparative investigation, such as theScottish Mental Surveys. Thus there wasvirtually no rise in scores between I932 andI947 in Scottish districts where children areunused to tests, but a rise of 3.2 points indistricts where children are relativelysophisticated.

I have surveyed some of the literature onpractice and coaching effects in my book,with Dr. Parry, on personnel selection (29).Important additional results have beenobtained by McIntyre (14) in Australia,Johri (ii) at Leeds, Dr. Watts (30) in Lon-don and Professor Peel (i8) at Birmingham;and a series of experiments is being carriedout by one of mystudents, Mr. D. T. Navathe.He is taking the precaution, which fewprevious investigators have done, of mixingtwo parallel versions of each test so thatthose pupils who are initially given version Atake version B as their final test, and thosewho are initially given version B take Afinally. Only in this way can slight differencesin the difficulties of the versions be con-trolled.The increase in score resulting merely from

the practice effect of taking one previoustest ranges from about 3 per cent to 8 percent. Peel compared I,593 pupils who tookMoray House Test 4I and Test 42 a fewweeks later, also I,367 pupils who took themin the reverse direction, and obtained anaverage rise of 3.I4 I.Q. points. Mr. Navathegot a rise of 4 per cent after two weeks usingsimilar tests. But both these sets of pupilsmay have had some previous familiarity withtests, and in another experiment at a privateschool where the boys were virginal the risewas 7.4 per cent. These, however, were boysof initial I.Q. around I20 and therefore per-haps more able than most to learn frompractice. Dearborn and Rothney (9), Rodger(20) and others have claimed that train-ability is greater at higher I.Q. levels; butone would naturally expect greater increasesbelow the average than above as a conse-quence of regression. Peel has attempted toallow for this regression effect and hasapplied his as yet unpublished technique toMoray House retest results of 7,000 pupils.Table IV derives from a smoothed graph

which I drew from his results. It shows thattrainability does indeed increase with intel-ligence up to a certain point, say I.Q. i2o,but thereafter there is a dropping off, notbecause of ceiling effect but because the verybright children understand the initial test sowell that they have little to learn.Navathe compared the improvement on

timed tests with that when ample time wasallowed; the latter amounts to 2.8 per cent,that is roughly three-quarters as great aswith the timed test. Hence the practice effectis not merely a matter of learning to makeefficient use of time or to read and follow theinstructions quickly.

TABLE IVSMOOTHED RESULTS (FROM PEEL) OF RETEST RISES

Initial I.Q. Mean RiseLevel on Retest140 2*9130 3.9120 4-I110 3.5100 3-290 2-68o I-970 (1.2)

Overall 3.14

With further practice between the initialand final test the improvement may be con-siderably greater. Thus Johri obtained anadditional increase averaging about I2 percent as a result of ten periods of practice atparallel tests on successive days. All inves-tigators, however, show that this tails off;in other words, that increased practice pro-duces diminishing returns.

TABLE VSUMMARY OF RECENT INVESTIGATIONS, IN STANDARD

SCORE (I.Q.) UNITSPractice CoachingEffect Effect

McIntyre 15-I8Johri ... I2'4 14.8Watts ... 5.4 9-5

... 3 32 8-2Navathe ... 4-o 8-I I

so ... T74 I6'2Under my fourth category, coaching or

instruction in how to answer group test itemsproduces very large rises. Watts obtainedan increase of 8 to 9 points from quite inten-sive coaching over ten weeks, but his testeeswere secondary modern pupils who alreadyhad some acquaintance with tests. Navathefound 8 to ii points in similar groups. But


in a private school two small groups whoreceived only one or three periods of coachingrose by i6 points. McIntyre's pupils hadfour and Johri's ten periods. It is, of course,difficult to compare these experiments sincethe tests and the age and previous sophistica-tion of the children varied. But it certainlyappears that an hour or two of coaching canbe as effective as much more prolongedtraining. Notice, too, that coaching is onlyabout twice as effective as the practiceresulting from a single retest. In Johri'sstudy ten coaching periods were slightlymore effective than ten practice periods, butto a statistically insignificant extent. Thus,from the standpoint of educational selection,the solution is obvious. If all primary schoolswere made to give a few hours' coachingshortly before the iI-year examination,previously unsophisticated children would bebrought up to the same level as those whohad already been coached; and schoolswhich coached for months beforehand would,in all probability, have been wasting theirtime. But the mere application of a shortpractice sheet or (as happens in some areas)of one preliminary test is clearly not suffi-cient to raise children to their effective limit.Many other points require investigation.

Different types of test differ in their suscep-tibility to coaching, but there is no clearevidence as to which are the most prone.Navathe found that a battery of separatelytimed subtests yields much the same improve-ment as an omnibus test of the Moray Housetype; also that allowing ample time reducesthe coaching effects only by some io-2o percent. He is trying now to find how lastingare the effects, also whether coached oruncoached test scores are the more valid. Onthe one hand we might expect many un-coached children to fail to manifest their trueability because they often do not understandwhat the test requires of them. On the otherhand, fully coached children may haveacquired what is mainly a familiarity withthe tricks of the tester and their scores maybe less valid indices of the intelligence theyshow at school or in daily life. How fartraining on one type of test transfers to othertypes is a matter on which the evidence is

confficting. In one experiment in theNavy (29), practice obtained by taking onebattery of tests produced a 3.6 per centimprovement on another quite differentbattery. In another recent investigation byMr. Elliott, five groups, totalling I,719 men,took the same set of five tests in five differentorders; and quite large variations, rangingup to 5.2 I.Q. points, were found when thesame test occurred in different positions.For example, the I.Q. score on an abstractiontest was 97.7 if taken first(or second, butIOI5 if it came later in the series.* On theother hand a straightforward mathematicaltest requiring creative responses appeared tobe upset if it followed other, selective-response, tests. Thus when taken first thescore was IO2, but after other tests 99.5.These tests all differed considerably from oneanother in form, and it seems likely that theset of responding to one type of item inter-fered with the sets required for other types.Thus most of the tests showed a negativepractice effect, i.e. more declines than riseswith lateness in the series.

Johri trained or practised his boys onanalogies, similarities, directions, reasoningand mixed sentences. He also applied, butdid not train them on, opposites, best answer,always has and absurdities. He claims that,of the latter, only absurdities showed anyimprovement; neither the practice nor coach-ing transferred to the other three. Navathegave four tests-verbal classification, verbal

TABLE VISUMMARY OF NAVATHE'S RESULTS

PointsPractice effect alone, on all four tests ... ... 4-oPractice effect on tests dissimilar to those

coached ... ... ... ... ... 5-oPractice effect on tests similar to those coached 4.ICoaching effect on tests specifically coached... 8.4Coaching effect when whole battery is coached I.ioanalogies, non-verbal classification and non-verbal analogies-and he trained four of hisgroups, one on each of these tests; othergroups were practised or coached on all tests.His results show that transfer effects were nogreater than those arising from simple prac-tice. Actually there was more transfer to

* The five groups were nearly, but not quite, identicalin initial ability. However, analysis of covarianceshowed the variations to be statistically significant.

RECENT INVESTIGATIONS OF INTELLIGENCE AND ITS MEASUREMENT I3I

dissimilar than to similar tests, though thedifference is insignificant. For example,those coached on verbal classification showedmore improvement at non-verbal analogiesthan they did either at verbal analogies or atnon-verbal classification. Furthermore, thosetrained for two periods on one test showedless improvement at this test, on the average,than those trained for the same total time onall four tests.

I would suggest the following conclusions,though I would like much more evidence.First, either practice or coaching has a rathergeneral effect in familiarizing testees withintelligence tests, inducing the set of workingquickly, of taking careful account of theinstructions, not wasting time on difficultquestions, being alert to tricky items and inimproving confidence and reducing anxiety.One might call this the sophisticated attitudeor even the morale of the testees. Dr. Wattssuggests that this attitude may not onlyimprove but also decline if testees are over-coached so that they get bored. Second,there is a highly specific coaching effectwhich helps only at the particular type ortypes of item coached or practised, and whichmay hinder facility in tackling other types.The more general sophistication may well

be stimulated to some extent by recentschooling or by the taking of any kind ofexamination. Adult recruits tend to showrather large practice effects, possibly becausetheir schooldays are so remote that they donot readily settle down to writing answers tosilly questions as quickly as possible. Butthis brings us to the fifth of my headings-the improvement in intelligence test per-formance attributable to environmental in-fluences and education. I shall not attemptto resummarize the earlier literature whichtended to show a definite, though quite small,environmental effect. Nor will I describe thenotorious Iowa studies, which claim to showvery large effects but which have beendevastatingly criticized by McNemar (I5).I would, however, like to discuss the startlingresults published in I946 by BernardineSchmidt (22).

B. G. Schmidt's Study ofEnvironmentalEffectsMiss Schmidt states that, during I935-42,

she took 320 Chicago children aged I2-I4,who had been diagnosed as feeble minded,all with I.Q.s below 70. Three groups, total-ling 252 adolescents, were educated for threeyears at special experimental schools wherethe curriculum was designed to increaseemotional and social adjustment, as well asto provide appropriate training of academicand manipulative skills and study habits.Another, control, group of sixty-eight caseswent to more conventional schools. Thefollow-up of subsequent careers continuedfor one and a half to four and a half yearsafter the end of schooling. Stanford-Binettests and certain educational tests and testsof personality adjustment were applied ateighteen-month intervals throughout. Themain findings regarding intelligence areshown in Fig. i. The controls show the com-

Fig. i

B. tCxMI1T?MVDTSR LTS

IIL ~ 4,to ItX/X

t aa 3 5 i n

monly observed drop over the four and ahalf years from an average I.Q. of 6o to 56.But the three experimental groups, startingat an average of about 52, rise to 73 in the

132 THE EUGENICS REVIEW

three years' schooling and continue to in-crease thereafter, Group I reaching 89 inseven and a half years. The Vineland SocialMaturity scale showed even greater improve-ments, and the personality test results wereconcordant. Apart from tests, half theexperimental children went on to ordinaryhigh school courses and 27 per cent graduatedfrom high school. When last surveyed, 83 percent were in regular employment and nearlytwo-thirds of these were in clerical or skilledwork. By contrast, the controls showed avery poor educational and employmentrecord.

Naturally these results have arouseddoubts and criticisms, but it has been verydifficult to find flaws, apart from some minorarithmetical slips. At least one reviewer (12)seems to suspect gerrymandering, but this isunlikely since the study is backed by severalhighly reputable psychologists. (I myselffeel grave doubts as to how children in theimbecile grade could have taken such per-sonality questionnaires as Bernreuter's.)The initial intelligence testing was all doneby trained psychologists of the Chicagoschools Child Study Bureau; thus we cannotobject that some of the children were diag-nosed as feeble minded because of bad test-ing, as occasionally occurs in Britain. Re-testings were, apparently, carried out byMiss Schmidt using i9i6 Stanford-Binetthroughout with Groups I, II and IV, andTerman-Merrill Form L throughout withGroup III. Moreover, an independent testeris said to have given a further test to some ofGroup I, getting virtually identical I.Q.s.

Nevertheless, certain weaknesses deservemention. First, there are always considerablefluctuations in I.Q.s during the teens. Thecorrelation between tests applied to a repre-sentative group four and a half to seven anda half years apart would hardly exceed .6o,and might be lower. Hence, due to regressioneffects alone, we would expect an apparentrise in average I.Q. of I9 points, that is from52 to 7I. The statistically untrained personfinds this very difficult to grasp, but it isreally quite simple. If a representative groupwas tested (with accurately standardizedtests) at I2 years and at i6 years there would

be as many children with I.Q.s below 70 onthe second as on the first occasion. But theywould not all be the same children. Owing toimperfect correlation, some of those scoringabove 70 at I2 fall below 70 at i6, and viceversa. If, therefore, we pick out only thosebelow 70 at I2 years (as Miss Schmidt hasdone) they inevitably show an average riseby i6; this rise has no psychological sig-nificance at all since we are neglecting theabove 70s who show a fall in the sameperiod.A second point is that the experimental

children were tested from five to eight times,the controls only twice. Schmidt denies thatthere is any practice effect when retests areas far apart as i8 months. I doubt this, par-ticularly with children near to their limit ofintellectual growth, and suspect that at least5 points of the average rise might be attri-buted to practice.

Thirdly, you will note that much thebiggest rise occurs in Group III, which wastested with Terman-Merrill. It amounts to35 I.Q. points in four and a half yearscompared with 26 points in Groups I and IIwho were tested with Stanford-Binet. Nowin spite of all Terman's care in standardiza-tion the new revision undoubtedly exag-gerates adolescent I.Q.s. Every psychologistwho has used Form L on children of I2 yearsupwards has been surprised by the numberof very high I.Q.s that turn up. In I940 Ipointed out that the vocabulary test alonewas considerably easier at the upper endthan the Stanford-Binet vocabulary, and Icarriedout a rough restandardization reportedin my book The Measurement ofA bilities (26).The first two columns of Table VII show thatchildren with a ten-year Terman-Merrillvocabulary score would score nine yearsonly on the Stanford-Binet or according tomy norms, while a twenty-year Terman-Merrill score corresponds to sixteen and ahalf years only. Next, over several years, mystudents at Glasgow University appliedTerman-Merrill to small groups of childrenalong with several other intelligence, educa-tional and performance tests believed to berelatively accurately standardized. In thisway I collected records for several hundred


children of various ages and, on graphingTerman-Merrill mental ages against othermental ages, found that Terman-Merrill wasprobably quite accurate around six to eightyears, but then gave more and more exag-gerated results. I combined these into arather conservative restandardization sum-marized in the third column of Table VII.Then in I949 the second Scottish MentalSurvey appeared, showing a tremendouspositive skew in the Terman-Merrill distribu-tion of eleven-year I.Q.s. Whereas 3 per centof the representative sample obtained I.Q.sbelow 70, not three but IO per cent obtainedI.Q.s above I30. It is quite simple to workout from the Scottish figures revised normswhich turn this skewed distribution into asymmetrical one, and an extract from thesenorms appears in the last column. My threeattempts at restandardization do not agreevery closely, but they all show the sametrend.* I do not know whether this inac-curacy in the Terman-Merrill scale arises

TABLE VIIRESTANDARDIZATIONS OF TERMAN-MERRILL MENTAL

AGESRevised M.A.s, Based on

Obtained Vocab. Glasgow ScottishM.A. Test Tests Survey

8 7:6 8:o 8:o10 9:0 9:7 9: IO12 10 4 II: I Ii:614 11 . 7 12: 7 I3: Ii6 14:1 14:3 14: 5I8 I5:3 15:11 15:820 I6:6 17 :7 I6: IO22 I7: 8 19: 3 17:9

from Terman's failing to secure representa-tive older groups, but I suspect that it is duemore to the fact that mental growth is noteven approximately linear after about tenyears. Several investigations suggest that itis logarithmic, i.e. that there is a gradualdeceleration. Terman makes some allowancefor this, but not enough. Hence, if successivemental age years above ten represent smallerand smaller increments of mental growth theexaggeration of I.Q.s that we have foundwould naturally follow. Psychologists arecoming more and more to the view that the

* Quite a close approximation to the average of thesefigures is obtained by counting each year above 7: 0as IO months. E.g.: Obtained M.A. 13: 10 = 7 0+82 months; Corrected M.A. = 7: o + IO/I2 x 82 =7: 0 + 68 months== I2:8.C

M.A. and I.Q. system of scoring tests isquite inadequate above ten years or so andare resorting to standard scores or percentiles(cf. Vernon (28)).

This long digression was necessary to bringout my third criticism of Miss Schmidt.Clearly a child of average intelligence atI2 years will obtain an I.Q. greater than IOOif retested with Terman-Merrill four or fiveyears later, merely because of this defect ofscaling. I calculate the increase as aboutI4 points. A bright child might easily showa much larger rise of 30 points. MissSchmidt's dull children would not be ex-pected to increase as much, because theywould still be below the level where Terman-.Merrill exaggerates most, but some 5 pointsof the rise in her Group III may well be due tothis. Whether similar exaggeration occurswith the old Stanford-Binet (as used inGroups I and II) we do not know. There isactually more evidence that it gives undulylow I.Q.s among older adolescents andadults. But everything depends on whatdivisor is used for calculating I.Q.s, and thisMiss Schmidt fails to tell us. If it was sixteenyears for those aged i6 or over, then thereported rise in average I.Q. is indeed sur-prising; if fourteen, then it is much less so.But in any case Schmidt's results are opento the criticism that the I.Q. standardizationof all the Binet scales is most unsatisfactoryfor older children and adults.* When we alsotake into account the natural effect of regres-sion and the probable practice effect, wemust admit that the claims for large increasesin I.Q. are very dubious. Personally I ammore impressed by the sociological data. Onecannot dismiss the finding that, when stepsare taken to improve the emotional adjust-ment and to provide a suitably stimulatingeducational environment, a large proportionof children once diagnosed as feeble mindedare converted into reasonably capable adults.I am sure, however, that the investigationneeds to be repeated with better controls andmore adequately scaled tests.

* Yet another likely reason for a spurious rise onTerman-Merrill is that the standard deviation of I.Q.sat I6-17 years is only three-quarters what it is at I2.I know of no information as to whether the same istrue of Stanford-Binet.

'34 THE EUGENICS REVIEW

Other Evidence of Effects of EducationFurther evidence of environmental effects

is forthcoming from studies of the intel-ligence of adults. Ever since large-scaletesting began in the American army in I9I7,two apparently conflicting results have beenobtained. On the one hand the average M.A.of representative adult groups has beenremarkably low; for example, the meanArmy Alpha test score corresponded to aM.A. of I3j. On the other hand, whenpupils or students in high schools or collegesare given intelligence tests of appropriatedifficulty they usually continue to showrising scores from 15 up to over 2o years.Hence it seems absurd to say that the aver-age individual reaches the lmits of hisintellectual growth at i3 to I4 years. Butthese two findings can be reconciled if weallow that growth continues as long aseducation or other intellectual stimulus con-tinues and that thereafter a decline sets in.The majority of the population, includingalmost all the average and dull adolescents,finish their education around 14 to i6, hencetheir intellectual growth and decline mightbe represented by lines A and B in Fig. 2,

Fig. 2

CUVESV OFMU,TAL CRRcWTI4AND WC

.~~~~~~~~~~-r~~~~~W0 2h

A Gs Cwhich reach their maximum height at about15 years. But a minority receive secondary or

college education to i8, 22 or later, and whenthis finishes they usually enter jobs wheretheir intelligence continues to be fully exer-cised. They may be represented by line C,which shows an increase till 25 or 30 before agradual decline sets in. Since, however, theyare relatively few in number, the mean M.A.of all adults around 20 years is even lowerthan the maximum that the average pupilreached at I5.Now what is the evidence for this hypo-

thesis ? Lorge (I3) retested a number ofadults at the age of 34 whose test results atI4 years were known. There was a cleareffect of education beyond I4. For example,adults who had received a university educa-tion were two years superior in mental ageto others who had had no further schoolingbut whose intelligence level had been thesame at I4. Similarly Terman and Oden (23)followed up their gifted California groupafter 2o years and gave them an individualsynonyms and analogies test. It may bededuced from their results that those whohad had the best education and had enteredprofessional careers scored best, relative totheir I.Q.s as children. Might these resultsnot be due, however, to the use of verbaltests with high educational content ? Actu-ally the same seems to hold for other typesof test. Thus the mean score for i8 yearrecruits on an abstraction test, which hasonly a small verbal component, is equivalentto that of I3 year children; but brighterrecruits score distinctly higher than brightI4-I5 year olds and dull recruits distinctlylower than dull school-leavers. AgainRaven (io) has published age norms from5 to 65 years for his vocabulary and for hisnon-verbal matrices tests, both of whichshow the tendencies pictured in my Fig. 2,namely the fanning out of ability duringadulthood and the earlier decline amonglower-grade adults.

Elsewhere (29) I have given the meanscores on the twenty minutes' progressivematrices for nearly go,ooo naval recruits,ranging in age from I7 to over 40. Theseindicate that an average decline has begunas early as i8 years, though it is difficult tobe certain of the comparability of the various


age-group samples. When the recruits areclassified by civilian occupation there is adefinite tendency (shown in Fig. 3) for such

Fig. 3

(20-M*,.)TAVI SC0M&

44

4342

.A-

40-

33.

3S'

34.30.21

2524

M4eAfM 26')zS 358A6R 4$ 4)A5S51OMI GRARM5 SHUAA 3)ECLWE OFw wIGeWrfl A&LW VWIJ OCCU?ATM CJW(ULi-

groups as mates and labourers, whose workdemands least intelligence, to drop morerapidly than clerks, electrical and wood-workers, who are likely to make more use oftheir brains in their jobs and leisure pursuits.A corollary of my hypothesis is that we

would expect the relation between intellig-ence and education to be closer in an un-selected adult population than the relationbetween intelligence and educational attain-ment among children still at school. So longas the educational stimulus is much the samefor all individuals, innate factors will be themain determiners of individual differences inabilities; but after I5 years the amount ofeducational stimulation varies widely, andthis, I would claim, affects the measuredintelligence. This is borne out by our factoranalysis, where the arithmetic test had ashigh a g loading as the non-verbal intellig-

ence test, and the latter had a substantialeducational loading. Hence, also, family-sizedifferences in our measure of g were prac-tically identical with differences in education.

ConclusionsNow to sum up the inferences that appear

to follow from these investigations: I acceptthe view of Burks (3, 4), of Professor Burt (5)and others, that, within a culturally fairlyhomogeneous group such as the primaryschool population of the U.S.A. or of GreatBritain, something like 75 per cent of thevariance in intelligence is attributable tohereditary factors, provided that care istaken to ensure a uniform degree of fami-liarity with the tests employed. When theamount of practice or coaching variesbetween different sections of such a popula-tion, no worthwhile conclusions at all can bedrawn. But I would say that this largehereditary influence is manifested onlybecause the environmental stimulation isfairly uniform for all children. They all hearmuch the same language and learn to usemuch the same concepts; they receive amore or less standardized education; theylive in a country where the same pictorial orother concrete symbols are current; they areall trained to attend to printed questions andto write down answers quickly, and so on.All these factors do of course vary betweensocial classes, between families within classesor between children within families, thoughnot enough to raise the environmental com-ponent of the I.Q. to more than 20-30 percent. They begin to vary more widely whenchildren are segregated into different types ofsecondary schools and after school-leavingage. Probably also they vary more widelyamong emotionally maladjusted childrenwho have difficulties in absorbing adult con-cepts, or among Gordon's canal-boat andgipsy children, or among such cases as MissSchmidt investigated who, through emotionalhandicap or intellectual weakness, had fallenby the wayside so that the education whichstimulates the growth of the ordinary childbecame too advanced for them. Certainlythe variations in concepts, habits of thought,and in attitudes to intellectual tasks, between


members of different nations or racesare so large that experts like Nadel (X7) andBiesheuvel (i) believe that a culturallyneutral intelligence test is an impossibilityand that psychologists who try to make racialcomparisons are wasting their time. Thisapplies just as much to performance or pic-torial or other non-verbal tests as to verbalones. But I am afraid that this conclusionholds also, though in lesser degree, for manyof the comparisons within relatively homo-geneous cultures in which eugenists areinterested. The stimulus value of the environ-ment in which the numerous children ofsocial-problem families on the one hand, andthe only children of university graduates onthe other hand, are reared, may account forno more than 25 per cent of the differences intheir observed intelligence; but surely this25 per cent is too large a component for thetest scores to be accepted as reliable measuresof genetic differences.

I do not think that I am saying anythingrevolutionary about intelligence tests. It is,after all, a commonplace of genetics thatqualities as such are not inherited, only thecapacity to develop these qualities underfavourable environmental conditions. ThusI would regard intelligence as the outcome ofthe interplay of innate potentiality and ofsuch conditions as good emotional adjust-ment and appropriate educational stimula-tion. I am not sympathetic to the argumentthat intelligence itself is an innate quality,and that our tests are inefficient at measuringit because they are subject to environmentalinfluences. Rather, I regard intelligenceoperationally as the general all-round abilitythat an individual manifests in his daily lifeadjustments, at school, in his job, or in testperformances, and all these manifestationsare environmentally as well as hereditarilydetermined. That brings me to my lastpoint: my criticisms of tests apply only totheir use in genetic studies and do not affecttheir application to practical problems suchas educational and vocational prediction.Success at school, or in a skilled job, ishelped by a superior environment or educa-tion and hindered by a poor one. Thus it isbetter predicted by a test which is culturally

biased than it would be by that hypotheticaland unattainable instrument-a pure test ofinborn ability.

REFERENCESi. Biesheuvel, S. Psychological tests and their applica-

tion to non-European peoples. 1949 Yearbook ofEducation. London: Evans Bros.

2. Bosanquet, B. S. The quality of the rural popula-tion. EUGEN. REV., I950, 42, 75-92.

3. Burks, B. S. Nature and nurture: Their influenceupon achievement. 27th Yearbook, NationalSociety for the Study of Education. Bloomington,Ill.: Public School Publishing Company. I928.

4. Burks, B. S. On the relative contributions of natureand nurture to average group differences inintelligence. Proc. Nat. Acad. Sci., 1938, 24,276-82.

5. Burt, C. L. Intelligence and Fertility. London:Eugenics Society and Hamish Hamilton. 1946.

6. Cane, V. R., and Heim, A. W. The effects ofrepeated retesting: III. Further experiments andgeneral conclusions. Quart. J. Exper. Psychol.,I950, 2, I82-97.

7. Cattell, R. B. Is national intelligence declining ?EUGEN. REV., I936, 28, I81-203.

8. Cattell, R. B. The fate of national intelligence:Test of a thirteen-year prediction. EUGEN. REV.,1950, 42, I36-48.

9. Dearborn, W. F., and Rothney, J. W. M. Predictingthe Child's Development. Cambridge, Mass.:Sci.-Art. 194I1.

IO. Foulds, G. A., and Raven, J. C. Normal changes inthe mental abilities of adults as age advances.J. Ment. Sci., 1948, 94, I33-42.

ii. Johri, S. R. The Effect of Coaching and Practiceon Intelligence Tests. M. Ed. Thesis, University ofLeeds. 1939.

12. Kirk, S. A. An evaluation of the study of BernardineG. Schmidt. Psychol. Bull., 1948, 45, 32I-33.

I3. Lorge, I. Schooling makes a difference. Teach.Coll. Rec., I945, 46, 483-92.

14. McIntyre, G. A. The Standardisation of IntelligenceTests in Australia. Australian Council for Educa-tional Research, Educ. Res. Ser., No. 54. Mel-bourne University Press. 1938.

15. McNemar, Q. A critical examination of theUniversity of Iowa studies of environmentalinfluences upon the I.Q. Psychol. Bull., 1940, 37,63-92.

i6. Martin, W. J. The Physique of Young Adult Males.Medical Research Council Memorandum No. 20.London: H.M. Stationery Office. I949.

17. Nadel, S. F. The application of intelligence tests inthe anthropological field. The Study of Society(edit. F. C. Bartlett and E. J. Lindgren). London:Kegan Paul. 1939.

I8. Peel, E. A. A Note on Practice Effects in Intellig-ence Tests. Brit. J. Educ. Psychol., 1951, 11.

ig. Penrose, L. S. The Galton Laboratory: Its workand aims. EUGEN. REv., 1949, 41, 17-27.

20. Rodger, A. G. The application of six group intel-ligence tests to the same children, and the effectsof practice. Brit. J. Educ. Psychol., 1936, 6,291-305.

21. Scottish Council for Research in Education and thePopulation Investigation Committee. The Trendof Scottish Intelligence. London: Utiversity ofLondon Press. I949.


22. Schmidt, B. G. Changes in personal, social andintellectual behaviour of children originally classi-fied as feebleminded. Psychol. Monogr., I946,60, No. 5.

23. Terman, L. M., and Oden, M. H. The Gifted ChildGrows Up. Stanford, Calif.: Stanford UniversityPreSs. 1947.

24. Thomson, G. H. Intelligence and fertility. EUGEN.REv., 1950, 41, I63-70.

25. Vernon, P. E. Intelligence test sophistication.Brit. J. Educ. Psychol., 1938, 8, 237-44.

26. Vernon, P. E. The Measurement of Abilities. Lon-don: University of London Press. 1940.

27. Vernon, P. E. Psychological studies of the mentalquality of the population. Brit. J. Educ. Psychol.,I950, 20, 35-42.

28. Vernon, P. E. The interpretation of intelligencetest results. To be published in Le Travail Humain,'95I.

29. Vernon, P. E., and Parry, J. B. Personnel Selectionin the British Forces. London: University ofLondon Press. I949.

30. Watts, A. F. Coaching for the Grammar SchoolEntrance Examination. Paper given at the Educa-tion Section, British Psychological Society, Dec.29th, 1950.

The American Journal of Human GeneticsA quarterly record of research, review, and bibliographic material relating to heredityin man, and to the applications of genetic principles in medicine, anthropology,psychology, and the social sciences.Edited for the AmmucAN SocrTY OF HuAN GENETICS by C. W. CoTRmAN, University of Michigan, incollaboration with C. N. HERNDON, M. T. MACRLiN, H. W. NORTON, BRONSON PRICE, N. F. WALKER

and A. S. WIENER.

Volume 3 CONTENTS Number 1SNYDER, L. H. Old and new pathways in human genetics.HERNDON, C. N., and JENNINGS, R. G. A twin-fanily study of susceptibility to poliomyelitis.BANTON, A. H. A genetic study of Mediterranean anaemia in Cyprus.KALLMANN, F. J., FEINGOLD, L., and BONDY, E. Comparative adaptational, social and psychometric data on the life

histories of senescent twin pairs.FINNEY, D. J. Review of: I. M. Lerner's Population Genetics and Animal Improvement.MOURANT, A. E. Review of: W. C. Boyd's Genetics and the Races of Man.WALKER, N. F. Review of: A. Lundstrom's Tooth Size and Occlusion in Tuins.McCULLOCH, C. Review of: A. Sorsby's Genetics in Ophthalmology.STEINBERG, A. G., and MULDER, D. W. Review of: C. H. Alstr6m's A Study of Epikpsy in its Clinical, Social and Genetic

Aspects.NEEL, J. V. Review of: A. B. Grobman's Our Atomic Heritage.BIBLIOGRAPHY OF HUMAN GENETICS, 1950, PART 3.

ANNUAL SUBSCRIPTION, $8.oo-SINGLE NUMBERS, $2.50

Subscriptions and inquiries regarding membership in the Society should be addressed to H. H. Strandskov,Department of Zoology, University of Chicago, Chicago 37, III.

Eugenics Review, Vol. XLI, No. 8

recent investigations of intelligence and its measurement

Documents